AI Models Struggle With Construction Blueprint Measurements

GlobeSt covers the results of Construction’s Last Exam (CLE), Togal.AI’s benchmark testing AI models on 33 real-world construction tasks. General-purpose AI models scored between roughly 2% and 34%, revealing significant challenges in accurately measuring and counting items on construction blueprints. The results highlight a key gap between AI’s performance on general tasks and the precision required for construction takeoffs, where small measurement errors can have real consequences for project bids. CLE provides a benchmark for evaluating how well AI can handle the technical demands of real-world construction work.